Study population and estimation of organ doses
The study population comprised 94,396 diagnostic radiation workers who enrolled in the NDR from 1996 to 2011, including radiologists (n = 1520), other physicians (n = 18,684), dentists (n = 15,705), dental hygienists (n = 13,488), radiologic technologists (n = 26,356), nurses (n = 7561), and other medical assistants (n = 11,082). The Korea Center for Disease Control and Prevention (KCDC) has been conducting monitoring programs for all radiologic technologists (both conventional and interventional) involved in diagnostic radiology since 1996. In addition, the KCDC maintains a centralized national dose registry and operates a lifelong follow-up management system for radiation dose . Medical workers involved with nuclear medicine and therapeutic departments are under the Nuclear Safety and Security Commission and were not included in this system. Registry information included name, sex, personal identification number, occupational group, quarterly dose data, and the beginning and end of the period of measurement. Dose measurements were collected quarterly by five personnel monitoring centers designated by the KCDC using a personal thermoluminescent dosimeter. Standard practice is for dosimeters to be worn under aprons at chest level. All instruments were calibrated annually.
Organ-specific doses were previously estimated for all diagnostic radiation workers after badge doses were calculated for workers exposed before 1996 . Briefly, the annual and cumulative individual badge doses based on Hp (10) (dose at a tissue depth of 10 mm from the dosimeter) were calculated by combining the quarterly badge readings for the workers enrolled in the NDR. Quarterly doses below 0.01 mSv, which is the lowest detectable level of NDR, were assigned at value of 0.005 mSv – the midpoint between 0.01 mSv and zero at the dose reconstruction. For workers who started working with radiation before 1996 (n = 13,178; 14.0% of the total enrollees in the NDR), historical badge doses were reconstructed using a model in which annual doses were determined as a log-linear function of time and age . The age at the time of first exposure was estimated for each sex and occupational group using the findings of our previous survey  to determine the first year of radiologic practice for individual workers. Then, organ doses were estimated by converting measured and reconstructed individual badge doses to each organ-specific dose and multiplying by two conversion coefficients provided by the International Commission on Radiological Protection: the organ-absorbed dose per unit of air kerma free-in-air and the personal dose equivalent per unit of air kerma free-in air [20, 21]. The organ dose methods was adjusted for the probability of apron use and the placement of the badge, and an attenuation rate of 0.8 was assumed for the use of a lead apron to reflect the shielding effect. The organs and tissues for which specific doses were estimated included the bladder, brain/central nervous system, breast, colon, esophagus, gallbladder, kidney, liver, lung, oral cavity and pharynx, ovary, pancreas, prostate, stomach, rectum, red bone marrow, thyroid, and uterus. The remainder included all solid cancers other than these solid tumors, excluding non-melanoma skin cancer.
For estimating realistic exposure scenarios of cancer risk, all workers were classified by sex, job start year, and occupational group. The estimated radiation doses were allocated to each group by year based on i) the reported NDR doses for the period of 1996–2011, ii) the reconstructed doses for the period before 1996, and iii) the 2011 NDR doses for the period after 2011 until 60 years of age based on the assumption that future radiation doses were the same dose as those in 2011. The age of 60 years was considered the average retirement age in South Korea (https://goo.gl/dtstFk), although the retirement ages varied slightly according to the occupational group. Radiation organ doses had right-skewed distribution; thus they were represented using two parameters of lognormal distribution - median and geometric standard deviations - to incorporate the dose uncertainty. After that, the age of first exposure according to sex and occupational group was imputed by accessing data from our previous survey on the average age of start of radiologic practice by diagnostic medical radiation workers . Three calendar years (1991, 2001, and 2011) of start of professional practice were selected to cover the full range of exposure scenarios in this population. Furthermore, 42 scenarios combining both sexes, three distinct years of start of professional practice, and seven occupational groups were constructed using NDR and survey data.
Prediction of cancer risk
Excess lifetime cancer risk was estimated for each scenario using the RadRAT risk assessment tool (RadRAT version 4.1.1), which was developed at the United States National Cancer Institute to estimate the lifetime risk of developing cancer from radiation exposure . The RadRAT program was developed based on the BEIR VII models  to estimate excess cancer risk from radiation exposure by incorporating current understanding about radiation-related risks with an allowance for uncertainties related to dose-response model parameters, minimum latency periods, dose and dose-rate effectiveness factor, and risk transport from population to population. The RadRAT program follows the BEIR VII assumptions by applying an uncertain dose and dose-rate effectiveness factor (DDREF) for all chronic exposures using a lognormal distribution with a geometric mean of 1.5 with no-threshold risk models. Input information required by RadRAT includes sex, year of birth, exposure history, and run-specific parameters . The RadRAT program can be freely accessed at https://irep.nci.nih.gov/radrat and has been applied in previous radiation-related cancer risk projection studies [7, 23,24,25]. However, to our knowledge, no study to date has applied the RadRAT program to medical radiation workers. The lifetime attributable risk (LAR), the probability of a premature incidence of cancer attributable to radiation exposure in a representative member of the population (i.e., the probability that an exposed population will develop a radiation-induced cancer during their lifetime), was calculated as excess cases per 100,000 by sex, occupational group, and first year of professional practice in this program. The lifetime baseline risk (LBR) for cancer, the cumulative baseline probability of having a cancer over the lifetime, was calculated based on South Korean cancer incidence rates in 2010. The lifetime fractional risk (LFR) – the ratio between LAR and LBR – was presented to express attributable risk relative to baseline risk; LFR is more stable than LAR regarding differences in population structure and cancer incidence rates .
The 42 exposure scenarios were applied to individual organs, 15 for male and 17 for female workers, producing a total of 672 lifetime exposure scenarios (i.e., seven occupational groups × three distinct years of start of professional practice × 15 organ sites for men plus seven occupational groups × three distinct years of start of professional practice × 17 organ sites for women). For each scenario, we input data on sex, year of birth calculated as the first year of work minus the age at the first year of work, and each organ dose considered as chronic exposure in the RadRAT program. Sex- and age-specific incidence rates in South Korea in 2010 were used for determining the baseline incidence rates and survival function was based on US general population 2000–2005 in the RadRAT program. Organ-specific lifetime cancer risk was estimated in each of the 672 exposure scenarios using the organ-specific median annual absorbed doses per job title at the specified age of first exposure beginning in three separate calendar years of work start (1991, 2001, and 2011) and ending at an assumed retirement at age 60 years. The LAR of all cancers combined was calculated as the sum of all risks of individual organs. The leukemia risk was estimated excluding chronic lymphocytic leukemia. The excess cancer risk was calculated with 90% uncertainty intervals to incorporate both statistical and subjective uncertainties computed by Monte Carlo simulations using the RadRAT program and the simulation sample size was 300. An example of the risk estimation process can be found in the Additional file 1: Figure S1.